Built by someone
just like you.

We did not want to go broke on compute.

We were building with AI. Seriously. Across products, across workflows, across things we were proud of. And then the bills arrived.

Not because we were being careless. Because every call, every task — formatting, summarising, extracting, classifying — was going to the same flagship model out of habit. We were paying Rolls-Royce prices to fetch the newspaper.

We knew smaller, faster models could handle most of what we were throwing at the big ones. We just had no reliable way to know which tasks those were, or to route them automatically without rebuilding our stack. So we kept overpaying.

We wanted the best answer, not the most expensive one.

Cutting costs without a routing brain just means using worse models for everything. That is not a solution. That is a different problem.

What we actually wanted was something that paid attention. Something that learned which model handled which type of task well — and sent each call to the right place. Not cheaper across the board. Smarter, call by call.

The cost reduction was a consequence of good routing. Not the target. The target was always: the best answer, reliably, at whatever that actually costs. It turns out that is significantly less than what we were paying.

We built MCP servers. We had no idea
who was using them.

We were building MCP tools. Exposing them. And then — nothing. No signal. No visibility. Were they being called? By which models? What tasks were they being attached to? Were they actually useful, or just sitting there costing us uptime?

And if we wanted to charge for access — to turn a useful tool into something sustainable — we had no mechanism. No credits, no metering, no way to say you've used this, here is what that cost.

We were flying blind over infrastructure we had built ourselves.

If we all build this separately,
what a waste.

Every AI team in the world is facing the same three problems. The compute bills. The routing question. The MCP blindspot. And most of them are either ignoring it or rebuilding the same scaffolding from scratch, quietly, inside their own org, never to be shared.

That is a profound waste of time and talent. But it is also something more important than that — it is a missed opportunity.

LLMs and MCPs are not just changing how software is built. They are changing how the world thinks, creates, decides, and works. That is not hyperbole. That is what is actually happening. And right now, the people navigating that shift are doing it alone, each in their own silo, solving the same problems in isolation.

There is another way. We can grab this moment as a people — not as individuals. Shared infrastructure. Shared intelligence about what models do what. A common layer that everyone benefits from, that gets smarter the more it is used. Not owned by one vendor. Not locked to one stack. A utility, like the internet itself.